What runs on your device when you click the button
This page uses the open source @imgly/background-removal library, a neural network that segments a photo into foreground and background and runs entirely inside the browser through WebAssembly. Nothing about your picture leaves the machine. The library is not loaded until you press Remove background for the first time; at that moment it fetches its model files, roughly 40 MB, and the status line shows the download percentage. The browser caches those files, so the second image on the same day, or a week later, starts processing immediately, and the tool keeps working with the network switched off.
Processing time depends on the hardware. A recent laptop typically finishes a 12 megapixel photo in 3 to 10 seconds; an older phone can take 30 seconds or more, and very large images may fail on devices with little free memory. Resizing a huge photo to 2000 pixels on the resize tool before cutting it out is a sensible shortcut and rarely costs visible quality in the mask.
What comes out
The result is always a PNG with a real alpha channel, saved under the original name with -no-bg appended. The preview sits on a checkerboard so you can see which parts became transparent. Pixels the model considers background are fully transparent; edges get partial transparency, which is what makes hair and soft outlines look natural when the cutout is placed on a new backdrop.
PNG with alpha is large. A 3000x4000 product shot that was a 2.5 MB JPG typically becomes an 8 to 15 MB PNG after background removal, because PNG is lossless and photographs do not compress well without loss. If you need the cutout on a website, convert it on the PNG to WebP page; WebP keeps transparency and usually brings the same file down to 300 to 600 KB.
Where the model does well and where it does not
People, pets, products on a table, cars and most objects with a clear outline against a distinct background come out clean on the first try. The model struggles in predictable places: thin strands of hair against a busy background, transparent or reflective objects like glasses and bottles, and subjects that share a colour with the background. It also has no idea what you consider the subject; in a group photo it keeps every person, and in a photo of a cup on a cluttered desk it may keep part of the desk. There is no brush to correct the mask on this page, so for those cases you get a starting point rather than a finished cutout.
One photo is processed at a time. Drop another file and the previous result is replaced, so download before moving on.
Practical uses and what to remember
Typical jobs are marketplace listings that need a white or neutral background, profile pictures, logos scanned from paper, and product images for a shop. For a listing, run the cutout, then place it on a plain background in any editor and export as JPG. For a sticker or an overlay, keep the PNG. The tool discards EXIF data along with the background, so the GPS position of a photo taken at home does not travel with the cutout.
Because the model file is served from a content delivery network, the very first run needs an internet connection even though your image is never transmitted. If you are on a strict corporate network that blocks the download, the status line will report a failure rather than silently uploading anything. For a fuller explanation of what in-browser processing means for privacy, see browser-based tools and privacy.